Rudas et al. (1994) [322]
Goodness-of-fit in the analysis of contingency tables. Model represented
as a two-point mixture where one component is the model of interest
$H$
(independence used in examples) and the other unrestricted. Index of fit
$\pi^{*}$
is smallest proportion of unresticted component, i.e.
proportion of data which cannot be described by $H$.
MLE of $\pi^{*}$
(by EM) and lower bound (profile likelihood). Also bivariate normal
example.
Clogg et al. (1995) [81]
Mixture index of Rudas et al. (1994)
[322] applied to mobility data, independence and
quasi-independence models in examples. Emphasis
(more than in [322]) on examining the `residuals',
i.e. the part of the data left to the unrestricted model. Comparisons
of these to the standard residuals and $\Delta$.
Xi and Lindsay (1996) [392] Estimating the mixture index of Rudas et al. (1994) [322] with sequential quadratic programming instead of EM.